This study presents a research project funded by the Health Effects Institute and conducted by Dr. Daniel Krewski of the McLaughlin Centre for Population Health Risk Assessment, University of Ottawa, Ontario, Canada, and his colleagues. It looks at the American Cancer Society Cancer Prevention Study II (CPS-II), a large ongoing prospective study of mortality in adults initiated in 1982. This study was one of two U.S. cohort studies central to the 1997 debate on the National Ambient Air Quality Standard (NAAQS) for fine particulate air pollution in the United States.

Air pollution is directly linked to death. In December 2020, a UK coroner ruled that air pollution was the cause of a fatal asthma attack that led to the 2013 death of nine-year-old Ella Adoo-Kissi Debrah who lived adjacent to a busy motorway (BBC News, 2022). The assignment of air pollution as the official cause of death on a death certificate was the first of its kind in the world (Reynolds, 2020). Though this was the first official assignment of air pollution as a cause of death, there are numerous studies linking air pollution exposure with mortality all over the world. Before the COVID-19 pandemic, the air pollutant PM 2.5 was identified as the "largest environmental risk factor in the United States" (Goodkind et al. 2019, p. 8780) and the cause of more annual premature deaths than traffic accidents and homicides combined (Goodkind et al. 2019). With the onset of the COVID-19 pandemic, researchers began assessing the impact of air pollution exposure on COVID-19 incidence and death. In a widely received, nationwide study linking air pollution exposure to COVID-19 mortality, Harvard T.H. Chan School of Public Health researchers, Wu et al., produced significant findings linking the impact of long term exposure to PM 2.5 to COVID-19 mortality across the contiguous United States. This 2020 study, published in ScienceAdvances, has been cited over 600 times, covered by 131 news outlets and downloaded over 15,000 times. Georeferenced data is routinely used in public health research such as this, however, the substantive influence of geography in the relationship between the treatment and outcome variable is often not considered in the model specifications, research design, nor the sampling strategy (Goldhagen et al., 2005; Matisziw, Grubesic, and Wei 2008). Additionally, the mechanism of data aggregation to an administrative unit may spatially misrepresent the data (Delmelle et al., 2022). As air pollution is a local, regional, and transboundary phenomenon (Nordenstam et. al, 1998; Goodkind, 2019), spatial autocorrelation, or spatially similar values, in the long term exposure to PM 2.5 among U.S. counties is likely. Despite the inclusion of maps indicating strong spatial trends in the long term exposure to PM 2.5 and COVID-19 mortality, the possible presence of spatial autocorrelation at the local level or spatial heterogeneity at the regional level was not investigated by the authors. Epidemiological studies invoking large, areal units may misrepresent the underlying, spatial processes of environmental health-hazards and produce unreliable treatment effect estimates when relating air pollution exposure to disease (Fotheringham and Wong, 1991; Kolak and Anselin, 2019). In this thesis, the fragility of the Wu et al. treatment effect estimate to unobserved confounding is assessed utilizing an alternative sensitivity analysis framework. This framework revealed that the estimate derived by Wu et al. (2020) is much more fragile to confounding than reported by the authors. Spatial analysis was then applied to investigate the possibility of spatial regimes (e.g. hotspots) in the treatment and outcome variables which may contribute to biased or inefficient treatment effect estimates. Strong levels of spatial autocorrelation and regional spatial heterogeneity in the long term exposure to PM 2.5, and to a lesser extent in the COVID-19 mortality rate, were confirmed by both computational and exploratory spatial data analysis. The highly variable associations between long term exposure to PM 2.5 and COVID-19 Mortality per U.S. Census Region or EPA Climatically Consistent Region delivered the expected result that the relationship between the treatment and outcome variable changes with changes in the sub-National definition of place. An understanding of the geography of the ubiquitous, locally variable and far-reaching PM 2.5, and its related health-hazard risks can contribute to an uncovering of the politics, power relations, and socioenvironments that coproduce differential access to clean air and the resulting uneven health burdens experienced by Black, LatinX, Asian-American, and immigrant communities. This is an essential step towards disentangling the relationships rendering clean air no longer an "open-access good" (V ron, 2006).

This concise, application-oriented text is designed to meet the needs of practitioners and students in applied fields in its coverage of major, updated methods in the analysis of survival data. Includes analysis of standardized mortality ratios, methods for proving attenuation of healthy worker effects, ordinal risk factors and other new areas of research. Timely and diverse case studies are presented, plus a complete data set on ESRD patients on hemodialysis. Moderate level of mathematics required.

This book highlights quantitative risk assessment and modeling methods for assessing health risks caused by air pollution, as well as characterizing and communicating remaining uncertainties. It shows how to apply modern data science, artificial intelligence and machine learning, causal analytics, mathematical modeling, and risk analysis to better quantify human health risks caused by environmental and occupational exposures to air pollutants. The adverse health effects that are caused by air pollution, and preventable by reducing it, instead of merely being statistically associated with exposure to air pollution (and with other many conditions, from cold weather to low income) have proved to be difficult to quantify with high precision and confidence, largely because correlation is not causation. This book shows how to use recent advances in causal analytics and risk analysis to determine more accurately how reducing exposures affects human health risks. Quantitative Risk Analysis of Air Pollution Health Effects is divided into three parts. Part I focuses mainly on quantitative simulation modelling of biological responses to exposures and resulting health risks. It considers occupational risks from asbestos and crystalline silica as examples, showing how dynamic simulation models can provide insights into more effective policies for protecting worker health. Part II examines limitations of regression models and the potential to instead apply machine learning, causal analysis, and Bayesian network learning methods for more accurate quantitative risk assessment, with applications to occupational risks from inhalation exposures. Finally, Part III examines applications to public health risks from air pollution, especially fine particulate matter (PM2.5) air pollution. The book applies freely available browser analytics software and data sets that allow readers to download data and carry out many of the analyses described, in addition to applying the techniques discussed to their own data. http://cox-associates.com:8899/